Transfer Learning for Diabetic Retinopathy Detection

Abstract

According to the International Diabetes Federation (IDF), there will be 552 million diabetics by 2034. The most common form of diabetic disease can affect the eyes, which called Diabetic Retinopathy (DR). It is a major factor in the development of blindness. Recently, Artificial Intelligence (AI) and deep learning (DL), two emerging computer science approaches, have boosted the possibility of detecting DR in its early phases, which means patient’s chances of losing their vision in the future will decrease. This paper presents a multilevel detection system for DR with different severity using transfer learning techniques. We used the APTOS2019 dataset from Kaggle, which contains retinal pictures, to conduct our experiments. Then, we use deep transfer learning technique with five models to generate the DR images features. The obtained results are very satisfactory in terms of accuracy.